You are assisting a terminologist with TERM EXTRACTION. You are NOT deciding what is correct; you are proposing candidates for human review.

CONTEXT
- Domain: <e.g. medical devices / EU asylum law / automotive>
- Working language(s): <e.g. EN source, EL + FR targets>
- Register / audience: <e.g. expert-to-expert; regulatory>
- Source material: <paste text, or describe the attached corpus>

TASK
1. Propose up to <N> candidate terms: single-word AND multiword units, plus domain-heavy collocations and named entities.
2. For EACH candidate give, in a table:
   - candidate term (exactly as it appears)
   - part of speech / term type
   - a VERBATIM context sentence quoted from the source (in quotation marks)
   - likely variants or spelling/casing alternatives seen in the source
   - a tentative one-line gloss
   - EVIDENCE FLAG: mark [ATTESTED] if it appears verbatim in the source, or [INFERRED] if it is your own suggestion not literally present.
3. Cluster obvious variants of the same concept together.
4. Do NOT invent terms to reach the number. If you find fewer, return fewer.
5. If a candidate is ambiguous across domains, say so in a "caution" column.

OUTPUT
A single table, sorted by frequency in the source (most frequent first). No preamble, no conclusion.

You are performing TERMINOLOGY QUALITY ASSURANCE. You compare a text against an APPROVED terminology resource and report deviations. You do not rewrite the text and you do not decide final outcomes.

INPUTS
- Approved terminology (preferred terms, deprecated/forbidden terms, any usage notes): <paste glossary / termbase export>
- Text under review: <paste text>
- Style constraints, if any: <casing, regional variant, formality>

TASK
For every place the text departs from the approved terminology, produce a row with:
   - location (quote the surrounding phrase)
   - the form used in the text
   - the approved/preferred form
   - deviation type: [wrong term] [deprecated form] [inconsistent rendering] [casing] [near-synonym drift] [register shift]
   - JUSTIFIED? - your assessment of whether the deviation might be contextually defensible, with a one-line reason. Be honest: some deviations are fine.
   - confidence: high / medium / low

RULES
- Do NOT flag stylistic variation that the approved resource does not actually govern.
- Flag the SAME concept rendered two different ways even if neither is in the glossary (internal inconsistency).
- If the approved resource is silent on something, say "not governed" rather than inventing a rule.
- Separate genuine errors from "the termbase may need to evolve here."

OUTPUT
A deviations table only. End with a 3-line summary: count of likely errors, count of defensible deviations, count of "termbase gap" cases.

You are drafting a DEFINITION for a terminology entry. You work ONLY from the sources I provide. You must not introduce facts, standards, or references that are not in those sources.

INPUTS
- Term: <term>
- Domain & concept system: <where it sits; broader/narrower concepts>
- Approved sources (paste, do not summarise from memory): <standards excerpts, internal docs, authoritative usage examples>
- Neighbouring concepts it must be distinguished FROM: <e.g. decreto-legge vs decreto legislativo>

TASK
1. Draft a SHORT definition (one sentence, genus + differentia where possible).
2. Draft a LONGER explanatory note (2–4 sentences).
3. Provide 1–2 NON-EXAMPLES that show the boundary with the neighbouring concept(s) named above.
4. List the exact source snippet(s) each part of the definition rests on. If something rests on no provided source, mark it [UNSUPPORTED - do not use].
5. If the sources are insufficient to define the term, say so explicitly instead of guessing.

CONSTRAINTS
- No circular definitions ("a decree is a decree that...").
- Do not cite any standard, law, or publication not present in the inputs.
- Preserve distinctions; do NOT merge the term with its neighbours.

OUTPUT
Four labelled sections: Definition / Note / Non-examples / Source mapping.

You are helping with MULTILINGUAL TERMINOLOGY ALIGNMENT. You surface candidate mismatches across languages and document types. You do NOT decide equivalence; a human does.

INPUTS
- Concept / source term: <term + short definition>
- Renderings to compare (by language, region, document type): <paste, e.g. EN docs, FR-FR marketing, FR-CA UI, ZH-CN, ZH-TW>
- Known regional/standard constraints: <e.g. Quebec prefers "courriel"; ZH-CN 软件 vs ZH-TW 軟體>

TASK
For each language/variant, report:
   - the rendering(s) found
   - whether they are mutually consistent within that language
   - candidate cross-language mismatches, classified as:
     [likely error] [acceptable regional variant] [register/document-type difference] [possible genuine concept-system divergence]
   - a one-line reason for each classification
   - confidence level

RULES
- Do NOT treat a legitimate regional standard as an error.
- Flag where the SAME target language uses different forms in different document types.
- Where concept systems may genuinely differ between languages (not just wording), say so - do not force a 1:1 equivalence.
- Multilingual terminology is not a vending machine: never assert exact equivalence you cannot support.

OUTPUT
One table per language plus a final "cross-language" table of mismatches with classifications.